GMKtec is launching the EVO-X3, a compact AI workstation built around AMD's high-performance Ryzen AI Max+ 395 APU. Early access registration opens on June 22, with the global launch set for June 29 at 12:00 AM PST. Despite a chassis small enough to sit on a desk, it supports up to 128GB of memory and an OCuLink port for connecting an external GPU, making it a machine clearly aimed at running AI locally. Sales will go through GMKtec's official channels.

A small workstation for running AI on your desk

The EVO-X3 packs enough processing power into a mini PC to handle generative AI and machine learning right at your desk. The configuration is squarely aimed at people who want to run large language models and similar workloads locally, without relying on cloud servers.

At its core is AMD's Ryzen AI Max+ 395. Known by the development codename "Strix Halo," this APU (a single chip that combines CPU and GPU) carries 16 Zen 5 CPU cores with 32 threads and boosts up to 5.1GHz. Graphics are handled by an integrated Radeon 8060S with 40 cores, and across the CPU, GPU, and NPU (a dedicated circuit for AI processing), the system claims up to 126 TOPS of AI compute. TOPS indicates how many operations (in trillions) can be processed per second; the higher the figure, the easier it is to handle heavy tasks like image generation and local AI inference.

The redesigned chassis shown this time was unveiled bearing the signature of AMD CEO Lisa Su, underscoring that this is a product built around AMD's flagship APU.

OCuLink support for external GPUs, with expansion in mind

One of the EVO-X3's defining features is an OCuLink port, which is rarely seen on mini PCs. OCuLink is a connection standard that exposes PCIe Gen4 x4 bandwidth externally, letting you attach an external graphics card to boost rendering or AI processing. Even for workloads where the integrated GPU falls short, the design leaves room to add an external GPU later and scale up performance.

Connectivity is comprehensive. It offers high-speed USB4 and HDMI 2.1, supports Wi-Fi 7 for wireless, and runs both Windows 11 Pro and Linux (Ubuntu). Because it drives the APU at high clocks inside a small body, power is said to peak at up to 140W, and the cooling system uses a triple-fan design to manage the heat.

Up to 128GB of memory and two SSD slots

The EVO-X3 supports up to 128GB of LPDDR5X memory. Since memory capacity dictates how large a language model can be handled, generous memory matters in a configuration built for local AI.

For storage, it provides two M.2 2280 PCIe 4.0 x4 slots, and GMKtec's materials list support for up to 16TB of total capacity. Because keeping several large AI models on hand quickly fills up storage, the two slots give plenty of headroom for expansion.

Price, availability, and a higher-end variant

GMKtec has set the global launch for June 29, but final pricing has not been officially confirmed at this point. Overseas media reports cite pre-launch discount prices of 3,600 USD (about 580,000 yen) for the 128GB memory / 2TB storage configuration and 3,849 USD (about 620,000 yen) for the 128GB / 4TB configuration. Since both are described as pre-release discounts, keep in mind that they may change at the official announcement.※1 USD = 161 JPY (as of June 19, 2026)

GMKtec also says it plans to release a higher-end variant later, powered by the "Ryzen AI Max+ PRO 495," with memory expanded to 192GB and up to 160GB of VRAM that can be allocated to AI processing. As an option for those who want to run even larger AI models locally, that variant is worth watching too.

Summary

The EVO-X3 is a compact workstation built squarely for running AI locally, combining the Ryzen AI Max+ 395 with up to 128GB of memory and an OCuLink port for adding an external GPU. Early access registration begins on June 22, and the global launch follows on June 29. Pricing isn't officially set yet, but reports point to a starting figure of 3,600 USD (about 580,000 yen), an aggressive price that reflects what a serious local-AI machine demands. With a 192GB variant also on the way, the EVO-X3 should help gauge just how far demand for running large AI models personally will spread.